CEREBELLAR TIMING SYSTEMS

Richard Ivry

Department of Psychology, University of California Berkeley, California 94720

I. A Modular Approach to CoordinationII. Cerebellar Contribution to Movement Timing

Ill. Perceptual Deficits in the Representation of Temporal InformationIV. Timing Requirements in Sensorimotor LearningV. Characterizing the Cerebellar Timing System

VI. Interpreting Cerebellar Activation in Neuroimaging Studies: A Challenge for the Tim­ing Hypothesis?

VII. ConclusionsReferences

Coordinated movement requires the normal operation of a number ofdifferent brain structures. Taking a modular perspective, it is argued thatthese structures provide unique computations that in concert producecoordinated behavior. The coordination problems ofpatients with cerebel­lar lesions can be understood as a problem in controlling and regulatingthe temporal patterns ofmovement. The timing capabilities of the cerebel­lum are not limited to the motor domain, but are utilized in perceptualtasks that require the precise representation of temporal information.Patients with cerebellar lesions are impaired in judging the duration of ashort auditory stimulus or the velocity of a moving visual stimulus. Thetiming hypothesis also provides a computational account of the role ofthe cerebellum in certain types oflearning. In particular, the cerebellumis essential for situations in which the animal must learn the temporalrelationship between successive events such as in eyeblink conditioning.Modeling and behavioral studies suggest that the cerebellar timing systemis best characterized as providing a near-infinite set of interval type timersrather than as a single clock with pacemaker or oscillatory properties.Thus, the cerebellum will be invoked whenever a task requires its timingfunction, but the exact neural elements that will be activated vary fromtask to task. The multiple-timer hypothesis suggests an alternative accountof neuroimaging results implicating the cerebellum in higher cognitiveprocesses. The activation may reflect the automatic preparation ofmultipleresponses rather than be associated with processes such as semantic analy­sis, error detection, attention shifting, or response selection.

INTERNATIONAL REVIEW OFNEUROBIOLOGY, VOL. 41

555 Copyrighr 0 HIlI7 by Academic Press,All rlghts of reproduction in any furm reserved,

0074-7742/97 $25.00

556 RICHARD IVRY

I. A Modular Approach toCoordination

The human brain can be described as an evolutionary device gearedto make our interactions with the world more efficient. Although we haveelaborate mechanisms for perceiving and learning about complex patterns,this information is only useful if we can respond to it in an appropriatemanner. Action is the ultimate goal of cognition, and action systems aredesigned to allow us to achieve our goals in a coordinated and flexiblemanner.

Given this, it is not surprising that so many parts of the brain areimplicated in motor control. A wide variety of neurological disorders candisrupt the production of coordinated behavior. In some of these disorders,such as apraxia, a loss of knowledge about the goal of behavior can beobserved (Heilman et al., 1981). However, in most movement disorders,the problem is a loss of coordination. The action may still be purposeful,but the control and execution of the action are disturbed. From this wecan create a list of the neural systems involved in coordination and skilledmovement. This list would include the motor cortex, the basal ganglia,various brain stem nuclei, and, of course, the cerebellum. But such a listwould only provide a description of the functional domain of a neuralstructure. It would tell us little about how a particular structure contributesto the overall computations required to achieve coordinated behavior.

Research since the mid-I980s has focused on developing a psychologicaland neural model of the components of coordination (see Helmuth andIvry, 1997). From this perspective, we would acknowledge that coordinatedmovement requires the normal operation of a number of different brainstructures. However, the emphasis would be on identifying the specificcontribution of these different structures. That is, we have worked from astarting assumption that there is a basic modularity to the organization ofthe motor system. Different neural structures contribute to movement byproviding distinct computations, the sum of which will determine whethera particular action is coordinated or not. This modularity notion has beenwidely applied in the realm of perception. It has not been as well advancedin the motor domain. This is not to say that many, or any, researcherswould argue that the basal ganglia and cerebellum perform the samefunction. Nonetheless, there has been a persistent tendency to describethe functional domain of motor structures in terms of tasks rather thancomputations. For example, the cerebellum may be described as essentialfor the production of well-learned movements whereas the cortex is essen­tial for the acquisition of new movement patterns, A modular perspective,however, might emphasize that both structures are involved in both skilled

CEREBELLAR TIMING SYSTEMS 557

and unskilled movements, but their relative contributions change in accordwith varying computational demands.

Figure 1 provides an overviewof some of the modules required for theperformance of sequential movements. For each neural structure shownon the left side, a component operation is listed on the right side. Forexample, this overviewcharacterizes the premotor cortex as playing a criti­cal role in movement selection (Deiber et al., 1991; Rizzolatti et al., 1990).Thus, ifone is to pick up a glass of water, the premotor areas will determinewhether that gesture is made with the right or left hand. The specificationofwhen that movement should occur and the fine tuning of the kinematicsof the particular gesture, however, are assigned to other neural structures.For example, the basal ganglia may playa critical role in the switchingfrom one action state to another (Hayes et al., 1995; Robertson and Flow­ers, 1990).

In this conceptualization, the cerebellum is proposed to playa criticalrole in establishing the temporal patterns of muscular activation. Thischapter reviewssome of the evidence that supports the idea that the cerebel­lum plays a unique role in representing temporal information. A centraltheme to be emphasized is that this computational capability may be ex­ploited in a variety of task domains. That is, the cerebellum can be viewedas an internal timing system that not only regulates the timing of muscularevents, but is also used whenever a precise representation of temporalinformation is required. This computational demand may arise in percep-

Brain Regjon

FrontalLobes

ParietalCortex

PremotorandSupplementaryMotorArea

•

•

Computations required for seqyentialmovement

GoalDevelopment

SpatialRepresentation and Planning

Movement Selection

Movement Specification and Initiation

Cerebellum

Basal Ganglia

MotorCortex

Temporal patterns of activation

Switching between differentpatterns

Movement Executjon

FIG. I. Overviewof modules required for performance of sequential movements.

558 RICHARt) IVRY

tion and learning, and as such, the cerebellum will he implicated in thesenonrnotor tasks. But this docs not mean that the essential function of thecerebellum has changed. Rather, the domain of cerebellar function hasbecome generalized because these other tasks utilize its timing capability(see Ivry, 1993).

II. Cerebellar Contn"bution to MowIrnent Tll1ling

While there has been much interest in nonmotor functions of thecerebellum, it is important not to lose sight of the lessons that have beengarnered from a century of neurological observation. The foremost signsof cerebellar dysfunction involve a loss of coordination (Holmes, 1939).The springboard for the timing hypothesis stems from consideration ofthe unique movement problems that result from cerebellar lesions.

Figure 2 depicts the electrornyographic (EMG) record associated witha series of movements produced by a patient with a unilateral cerebellarlesion (Hore etal., 1991; see also Hallett et al., 1975). As would be expectedwith this pathology, the patient's problem were restricted to the ipsilesionalside, and thus the nomal records are from movements produced by the same

A Elbow B Wrist c Finger

Pos

Vel

Ace

AgNormal

Ant

AgCerebellar

Ant

cerebellarn -==-­---r- normal

~

,o

i

0.6s,o

i

O.es,o

i

0.6s

FIG. 2. Kinematic (rop) and EMG (bottom) records from ~ingll':iointmovements producedby a patient with the contralesional hand (Normal) or ipsilesional hand (Cerebellar). FromBore Pol nl. (1991), with permission.

CEREBEUAR TIMING S'\'STEMS 559

patient when using muscles on the contralesional side. On the unimpaired,normal side, the EMG record shows a biphasic pattern, with the antagonistmuscle becoming active near the peak of the agonist activity.The antagonistprovides the necessary braking force that will allow the movement to termi­nate at the target location. A different pattern emerges when we look atperformance on the impaired side. Here the onset of the antagonist isdelayed and fails to brake the movement. The patient ends up being hyper­metric, i.e., overshooting the target. In addition, there is an intentiontremor as the person hones in on the final goal of the movement, with thedelayed antagonist activity producing a series of overshoots. Thus, thedisruption of the temporal pattern of muscular events leads to both hyper­metria and intention tremor. Other work has indicated that the problemis primarily in the timing of the muscle patterns. For example, these patientscan scale the agonist burst when producing movements of different ampli­tudes (Hore etal., 1991). Thus, while the timing hypothesis does not exclu­sivelyaccount for these results, it does meet the basic criterion of providinga consistent account of the coordination problems observed following cere­bellar pathology.

In our research, we have looked for more direct evidence ofa cerebellarinvolvement in timing. We began with a simple motor task in which patientswith a variety of neurological disorders were tested on a timed tapping task(lvry and Keele, 1989). Each trial began with a synchronization phase inwhich the subject tapped along with a series of computer-generated tonesseparated by 550 msec. After about 6 sec, the tones were terminated andthe subject was instructed to continuing tapping, trying to maintain thetarget pace. Tapping continued until the subject had produced 30 unpacedintervals. Each subject completed at least 12 trials in this manner.

Overall, the patients tend to approximate the target interval in a consis-­tent manner. The primary focus was on the standard deviation of theintertap intervals. Patients with either cerebellar or cortical lesions weremore variable than control subjects on this task (lvry and Keele, 1989).Although this result is not surprising given that all of the patients wereselected because of their motor problems, this crude measure still provedsufficient to differentiate between the patient groups. In particular, patientswith Parkinson's disease, a disorder of the basal ganglia, performed compa­rably to the control subjects.

There are many reasons why an individual may be inconsistent on thetimed tapping task. Variability would, of course, be inflated if an internaltiming system was damaged, creating noise in a process determining wheneach response should be initiated. However, a central timer might be intact,but its commands may be inconsistently executed due to problems in themotor implementation system. Wing and Kristofferson (1973) developed

560 RICHARD IVRY

a formal model which partitions the total variability observed on this tappingtask into twocomponent parts. One component is associated with variabilityin central control processes including an internal timer. The second compo­nent is associated with variability arising from implementation processes.A description of this model and empirical justification for its primary as­sumptions can be found in Wing (1980). Ivry and Keele (1989; also Ivry etal., 1988) describe neurological evidence in support of the model.

The Wing and Kristofferson model was used to analyze in detail theperformance of patients with unilateral cerebellar lesions (Ivry etal., 1988).This group waschosen because the patients could serve as their own control,i.e., their performance could be compared with the impaired, ipsilesionalhand against that of their unimpaired, contralesional hand. In this analysis,the cerebellar group wasseparated into those with medial lesions and thosewith lateral lesions. Motivation came from consideration of the anatomyof the cerebellum (see Chez, 1991). The output from the lateral regionsis primarily ascending, ending up in the motor and premotor cortex. Theoutput from the medial regions is primarily descending, ending up in brainstem nuclei or synapsing directly on spinal circuits.

The results showed a double dissociation (Fig. 3). When tapping withtheir ipsilesional hand. the increased variability in patients with laterallesions was attributed to the central component (lvry et al., 1988). Incontrast, the increased variability for the patients with medial lesions wasattributed to the implementation component. This dissociation, coupled

MedialCerebellar Patients40

35

30(/)

.§.25

~ 20

Ii 15CiS

10

5

o

lateral Cerebellar Patients

Clock Clock Implementation

FIG.3. Estimates ofclock (central) and implementation variabiliryo» the repetitive tappingtask. Hatched bars, unimpaired hand; solid bars. impaired hand. Modified from Ivry et at.(19Hl:l) .

CEREBEllAR TIMING S\STEMS 561

with the following perception results, led to the proposal that the lateralcerebellum plays a critical role in controlling the timing of these periodicmovements. This does not exclude the possibility that the medial cerebel­lum is also involved in timing. Its contribution to coordination may alsobe time based, but in a manner that anticipates and corrects ongoingmovements (e.g., efference copy) rather than one that initiates new motorcommands (see Keele and Ivry, 1991).

",. Perceplual Deficits inthe Represenlalion 01 TemporaIlnfonnaIion

The domain of the cerebellar timing system extends beyond motorcontrol. One line ofsupport for this hypothesis is that patients with cerebel­lar lesions are also impaired on perceptual tasks that require precise timing.This work was motivated by correlational studies showing that a commontiming system was used in motor and perceptual timing tasks (Keele et al.,1985; Ivry and Hazeltine, 1995). In our patient work, we employed a sim­ple duration discrimination task (lvry and Keele, 1989). On each trial,two pairs of two tones were presented. The first pair was separated by400 msec; this provided a standard, reference interval. The second pair oftones formed an interval that was either shorter or longer than 400 msec.The subject made a two-alternative forced choice response. An adaptivepsychophysical procedure was used to determine the difference thresholdrequired for each subject to be accurate on approximately 72% of the trials(Pentland, 1980). For a control task, a similar stimulus configuration wasused, but here the intensity of the second pair of tones was varied. Thesubject judged if the second pair was softer or louder than the first pair,and the same adaptive procedure was used to determine the differencethreshold for loudness perception.

The results in this study provided a second double dissociation (Fig.4). Only the patients with cerebellar lesions were impaired on the durationdiscrimination task (lvry and Keele, 1989). Patients with Parkinson's diseasewere as accurate as controls and. more importantly, so were the corticalpatients. In fact, the latter group wasfound to be impaired on the loudnessdiscrimination task, perhaps because some of the lesions extended intothe temporal lobe. While we were not particularly interested in the neuralbasis of loudness perception, the fact that the cortical group was selectivelyimpaired on this task provides further weight to the claim that the cerebellardeficit on the duration discrimination task reflected a specific deficit intime perception.

562 RICHARD IVRY

Time Perception Task50

45

40

u; 35§.

30:>

! 25'0C.!! 20CIJ

15

10

5

0Controls Parkinson Cerebellar Cortical

1.8Loudness Perception Task

1.6

1.4

co 1.2Eo:>

!'0c 0.8asli)

0.6

0.4

0.2

0Controls Parkinson C6rebenar Cortical

Group

Ftc. 4. Perceptual acuity on two psychophysical tasks, duration (top) and loudness (bot­tom), for lour groups of age-matched subjects (Stand Dev, standard deviation). Modifiedfrom Ivryand Keele (1989).

Converging evidence for a role of the cerebellum in time perceptioncomes from a positron emission tomography (PET) study (jueptner et al.,1995). The experimental task in this study was essentially the same as that

CEREBEllAR TIMING SYSTEMS 563

used in the patient study of duration discrimination. For their control task,the subjects simply listened to the stimuli and made alternating fingerresponses to control for motor output. Significant increases in blood flowwere observed bilaterally, with the foci centered in the superior regions ofthe cerebellar hemispheres.

Patients with cerebellar lesions are also impaired in their ability to judgethe velocity of a moving visual stimulus (Ivry and Diener, 1991). In thesestudies, the subjects viewed displays consisting of a series of dots that sweptacross the screen. As a dot reached the end of one side of the screen, anew dot appeared at the other end. This configuration was adopted tominimize tracking eye movements. The subjects were required to judge inwhich of two successive displays the dots moved fastest. A control task useda similar procedure, but here the location of the dots was adjusted in thevertical plane and the perceptual judgment was position based. The patientswere significantly impaired only on the velocity task. This finding has beenreplicated in another laboratory (Nawrot and Rizzo, 1995), and similarvelocity perception deficits have been found with somatosensory stimuli(Grill et al., 1994).

Ivry and Diener (1991) hypothesized that a faulty representation of thevelocity of a moving stimulus, a time-based computation, may underliesome of the occulomotor problems observed in these patients (see Leighand Zee, 1991). For example, in order to generate an appropriate saccade,it is necessary to have an accurate representation of the future position ofa moving stimulus. By measuring eye movements, Ivry and Diener (1991)showed that the perceptual deficit was not an indirect consequence of amotor problem. Patients who were able to maintain fixation were as im­paired as those who were unable to suppress intrusive eye movements.

IV. TIming Requirements inSensorimoIar Learning

Summarizing to this point, we have consistently observed impairmentsin patients with cerebellar lesions on tasks designed to require precisetemporal processing. The timing hypothesis not only provides an accountof the motor problems faced by these patients, but also leads to predictedperceptual deficits. A further source of evidence for this hypothesis comesfrom a very different paradigm: research showing that the cerebellum isinvolved in certain types of sensorimotor learning. This work also under­scores the usefulness of thinking about brain structures in terms of theircomponent operations rather than in terms of their task domains.

564 RICHARD IVRY

A large literature has been assembled since the mid-1980s demonstra­ting that the cerebellum plays a critical role in eyeblink conditioning (forreviews, see Thompson, 1990; Yeo, 1991, this volume; Woodruff-Pak, thisvolume). In the standard eyeblink conditioning paradigm. a tone is usedas the conditioned stimulus and precedes an airpuff to the eye by a fixedinterval such as 400 msec, Although the animal will make an unconditionedresponse to the airpuff, learning centers on the fact that, over time, theanimal comes to make a conditioned response in anticipation of the airpulT.

A number of studies have demonstrated that animals with cerebellarlesions fail to learn the conditional response. Moreover, learned responsesmay be abolished following cerebellar lesions. The deficit does not appearto be a motor problem in that the same animals continue to produce theunconditioned response. Similar results have been reported in humanliterature. Patients with bilateral cerebellar lesions show a severe impair­ment in eyeblink conditioning (Daum et al., 1993; Topka et al., 1993).Patients with unilateral lesions are more severely disrupted on the sideipsilesional to the lesion (Woodruff-Pak et al., 1996).

Much of the research with this paradigm has focused on identifyingthe neural circuitry that is critical for this simple form of learning. Ourinterest in this phenomenon centers on the computational characteristicsof eyeblink conditioning. In particular, one critical aspect is the need fora precise representation of the temporal interval between the conditioningstimulus and the unconditioned stimulus. The animal learns to make ananticipatory conditioned response. The fact that the response is anticipatoryis what makes it adaptive: by blinking before the airpuff, the animal is ableto attenuate the aversive consequences of the airpuff. Equally important,the conditioned eyeblink must be appropriately timed. The animal shouldnot blink too soon or the blink may be finished before the airpuff isdelivered. Thus, this paradigm demands that the animal learn not only toassociate the two stimuli, but learn the precise temporal relationship be­tween the tone and the airpuff. The evidence that they do just this is shownby the fact that the timing of the learned response is alwaysjust prior tothe airpuff, regardless of the interstimulus interval (Kehoe et al., 1993;Wickens et al., 1969).

Thus, it could be argued that the cerebellum is not essential for eyeblinkconditioning because ofsome general role in classical conditioning. Rather,the essential reason is because this type of learning requires precise timingand the cerebellum is uniquely suited for providing this type of computa­tion. Classical conditioning of other responses that do not show the sametemporal constraints do not involve the cerebellum (Lavond et al., 1984).

Two other points are relevant for the extension of the timing hypothesisto classical conditioning. First, it is of interest to note that at least four

CEREBELLAR TIMING SYSTEMS 565

computational models of eyeblink conditioning have been proposed sincethe early 1990s (Bartha et al., 1992; Buonamano and Mauk, 1994; Desmondand Moore, 1988; Grossberg and Schmajuk, 1989). A central feature of allfour models is that they contain mechanisms which can provide an explicitrepresentation of temporal information. This feature has not been part ofneural models developed for other task domains.

Second, Perrett et al. (1993) have observed an important dissociationbetween the effects of lesions of the deep cerebellar nuclei and lesions ofthe cerebellar cortex on the conditioned response. Nuclear lesions abolishthis learned response, presumably because these lesions destroy all of theoutput from the cerebellum. In contrast, cerebellar cortical lesions do notabolish the conditioned response. Rather, they disrupt the timing of theresponses, with many of the eyeblinks occurring shortly after the onset ofthe tone. It is as if the delay imposed by the cortex to ensure that theeyeblink occurs at the right point in time is abolished. In this situation,the response is, of course, no longer adaptive.

v. Characlerizing the Cerebellar Tnning System

The timing hypothesis provides a general description ofcerebellar func­tion. This specifies a unique computational role of the cerebellum that isnot limited to motor control, but also can account for perceptual andlearning deficits associated with cerebellar lesions. An important question,of course, is how to best characterize the timing properties of the cerebel­lum. When we think ofa timing system such as a clock, our first inclinationis to think about oscillatory processes such as a pacemaker. Indeed, mostmodels ofinternal timing systems center on a clock-counter system in whichoutputs from an endogenous oscillator are stored in a counter mechanism.The full models also include various memory and decision processes, aswell as a gating process that can control whether the periodic outputs ofthe dock are stored (e.g., Gibbon and Church, 1990).

For the most part, these models have been developed to account forbehaviors that span intervals considerably longer than those studied inmotor and perceptual tasks, usually on the order of at least several seconds.A general assumption in the timing literature is that these same mechanismswould apply for millisecond timing. The basic idea of a pacemaker as aperiodic process, however, may not provide the best description of thecerebellar timing system. Rather, the cerebellum may be viewed as providinga near-infinite set of hourglass or interval-type timers (see Ivry and Hazel­tine, 1995). Each hourglass represents a particular interval. There may be

566 RICHARD IVRY

some sort of organization to these units. a chronotopic map within thecerebellum. However, computational models such as that offered by Buena­Olano and Mauk (1994) capture the notion of multiple timers. but with adistributed representation that can be shaped as a function of the temporaldemands of a particular task.

In our initial tapping studies. each patient served as his or her owncontrol. The tapping performance was compared between trials in whichthe patients used their contralesional, unimpaired hand with trials in whichthey used their ipsilesional, impaired hand. Patients with lateral lesionswere found to have higher clock variability on the impaired side. Thissuggested that there are at least two clocks: a damaged one on the Iesionedside and an intact one on the normal side.

More recently. the notion of multiple timers has been explored in avariant of the repetitive tapping task. Franz et al. (l996b) examined whatwould happen when patients with unilateral hemispheric lesions tappedwith both hands simultaneously. The results were quite surprising (Fig. 5).In the unilateral condition, our original findings were replicated. Variabilitywas higher when tapping with the ipsilesional hand. and when the Wing­Kristofferson model wasapplied, the difference wasattributed to the centralcomponent. However. this difference disappeared in the bimanual condi­tion. Now the two hands were equally consistent. Most interesting, the badhand became better. We were puzzled as to how to interpret the results.One possibility was that the patient could somehow rely on the good timer.Perhaps it provided a more salient signal and thus dominated performance.

30Central Variability Implementation Variability

30

25 25

20 20f/lE 15 15

10 10

5 5

0 0Unimpaired Impaired Unimpaired Impaired

Hand Hand Hand Hand

Fie, 5. Estimates of central and implementation variability in cerebellar patients duringeither unimanual (solid bars) or bimanual (hatched han) repetitive tapping. Estimates arealwaysof within-hand variability and are plotted for both unimpaired. contralesional handsand impaired, ipsilesional hands. Modified from Franz et at. (l!m6IJl.

CEREBEllAR TIMING SYSTEMS 567

However, this position was abandoned when the experiment was re­peated in healthy subjects (Helmuth and Ivry, 1996). Thirty right-handedcollege students were asked to tap with their right hand, their left hand,or both hands. The results, in terms of total variability, are shown in Fig.6. Two points stand out. First, subjects were slightly more consistent whentapping with their dominant hand. This effect was linked to higher imple­mentation variability in the nondominant hand (see also Sergent et al.,1993). Second, and more important, timing variability was reduced whensubjects tapped with both hands at the same time, i.e., each hand becamemore consistent when the two hands moved together. As with data fromcerebellar patients, the Wing and Kristofferson (1973) model attributedthis bimanual advantage to reduced variability in central control processes.

These results argue against the hypothesis that performance is deter­mined by a single timer in bimanual tapping. If this were the case, wewould not expect to see an improvement in the control subjects for bothhands. Given these results, Helmuth and Ivry (1996) considered an alterna­tive model to account for the bimanual advantage. This model centers ona simple, yet counterintuitive hypothesis. Specifically, Helmuth and Ivry(1996) postulate that there are two independent timers during bimanualmovements: one associated with movements of the right hand and a secondassociated with movements of the left hand. The bimanual advantageemerges because of a central bottleneck that limits when central motorcommands can be issued, a mechanism believed to underlie the ubiquitous

25

u; 20

.s~ 150c:

111rn 10

5

0Left Hand Right Hand

FIG. 6. Total variability in unimanual (solid bars) and bimanual (hatched bars) conditionsin healthy subjects on the repetitive tapping task (Stand Dev, standard deviation). Modifiedfrom Helmuth and Ivry (1996).

568 RICHARD IVRY

temporal coupling observed in multi-effector actions. The authors proposethat this bottleneck must integrate the two timing signals and issue a singlecommand. The bimanual advantage results as a statistical consequence ofthis integration process (see Helmuth and Ivry, 1996). Thus, by this account,there is not a single timer in bimanual movements, but separate timers thatarc integrated hy a constraint in terms ofwhen central motor commands canbe implemented.

As noted earlier, the timing of ipsilesional movements produced bycerebellar patients becomes lessvariable when accompanied bycorrespond­ing movements of the contralesional hand. Moreover, during bimanualmovements, the patients show strong interlimb coupling. These two resultssuggest that the temporal integration process producing the bimanualadvantage is not dependent on the cerebellum. Timing and temporal cou­pling appear to be associated with different neural systems (Franz et al.,1996a).

In other work with normal subjects, a similar reduction in variabilitywas found when subjects made simultaneous finger and foot movementsregardless of whether the two limbs were on the same or different sides ofthe hody (Helmuth and Ivry, manuscript in preparation). The generalityof this effect is in accord with the notion that the cerebellum is bestconceptualized as an array of a near-infinite set of timers. As long as themovements invoke nonoverlapping neural elements, the bimanual advan­tage will be obtained. In this sense, timing is assumed to reflect a generaland unique computational capability of the cerebellum. This capability willhe exploited whenever a task requires the timing function of the cerebel­lum, but the exact neural elements that will be activated will vary from taskto task.

VI. Interpreting cerebellar Activation in Neuroimaging Studies: AChallenge forthenming Hypothesis?

The idea ofmultiple timers has potential implications beyond providinga characterization of the cerebellar timing system. It can also lead to anovel perspective on recent functional neuroimaging evidence that pointsto a role for the cerebellum in cognition. In these studies, the cerebellumis activated even when the experimental and control tasks are equated interms of their motor requirements (e.g., Jenkins et al., 1994; Kim et al.•1994; Petersen et al., 1988; Raichle et al., 1994).

What do the metabolic events seen in PET and functional magneticresonance imaging studies reflect? In the neuroimaging studies, there is a

CEREBELLAR TIMING SYSTEMS 569

common denominator across those conditions that produce significantincreases in metabolic activity in the cerebellum. This common denomina­tor is that these conditions are invariably more difficult than the comparisonconditions. That is, there seems to be a strong correlation between taskdifficulty and cerebellar activation. One operational definition of "taskdifficulty" would be to determine the possible set of responses. By thisdefinition, difficult tasks are those associated with more response alterna­tives.

A review of the imaging literature indicates that experiments demon­strating a role of the cerebellum in cognition confound the experimentaland control tasks in terms of the number of response alternatives. Considerthe seminal language study of Petersen et al. (1988). In the two criticalconditions, the stimuli were identical, the presentation of a single concretenoun. In the control, repeat condition, the subjects simply read the word.As such, there was only a single possible response. In the experimental,generate condition, the subjects had to name a verb that was a semanticassociate of the stimulus. Here, we would expect there to be many possibleresponses, at least in the first trial. For example, if the target word was"apple," possible responses in the generate condition would be "eat,""peel," "throw," and "boot up."

The fact that the cerebellum was more active in the generate condition,even though both conditions required the subjects to articulate a singleword, is frequently cited as demonstrating a cognitive role for the cerebel­lum.' However, an alternative, essentially motoric view would be that thecerebellar activation reflects the preparation ofall of the possible responses.By this logic, the increased activation in the generate condition resultsfrom the fact that there are more potential responses and that the cerebel­lum does its part to prepare for each one.

Is this cognition? Would the fact that the cerebellum prepares all possi­ble movements imply a cognitive role for this structure? On the other hand,we might imagine a motor theory of cognition in which the choice aboutwhich response to make requires the ability to plan that response. As such,the cerebellar contribution would seem to be cognitive. On the other hand,the cerebellum may be viewed as a system that simply goes about thebusiness ofpreparing its contribution for candidate responses. The cerebel­lum could be entirely unrelated to the more cognitive aspects of the task.

I Fiez et al. (1992) have provided converging evidence based on a case report of apatient with a cerebellar lesion. This patient had great difficulty on a variety of semanticassociate tasks. despite his superior performance on standard neuropsychological tests. How­ever. Helmuth etal. (1997) tailed to find similar deficits in a group study ofcerebellar patients.The reason lor this discrepancy remains unclear.

570 RICHARD IVRY

It may simply be a slave system in the sense that for any possible responsegenerated in the cortex, the cerebellum helps prepare to make that re­sponse. Some responses will be selected, but the cerebellum need not bepart of this more cognitive, decision-making process.

If this hypothesis is correct, the imaging results need not be at oddswith the hypothesis that the primary function of the cerebellum involvestiming. Perhaps the cerebellum faithfully goes about preparing the tempo­ral patterning of the movements associated wit.h all of the possible re­sponses. The idea that each response requires its individual cerebellaractivation is in accord with the hypothesis of Helmuth and Ivry (1996)regarding bimanual movements. In that work, the authors proposed thatindependent timers were invoked for each effector, even producing syn­chronous movements. In this reinterpretation of the imaging data, theauthors propose that nonoverlapping motor plans are prepared for allcandidate responses.

Raichle et al. (1994) have reported that, with practice, the cerebellaractivation in the generate task diminishes. Given that subject'! report thesame semantic associate on successive trials, it would be expected that thenumber of potential responses also becomes reduced, eventually equal tothat of the repeat condition (i.e., are possible response). Reductions incerebellar activation with practice have been observed in other PET studies(Friston el al., 1992). These results are consistent with the idea that oneaspect of skill automatization involves constraining the number of possi­ble responses.

VII. Conclusions

Three main points emerge from this chapter. First. the cerebellum ispart of a distributed system for motor control, and it is necessary to identifythe component operations of the different structures involved in motorcontrol. The timing hypothesis provides a specific functional role for theunique contribution of the cerebellum.

Second, this timing capability appears to extend beyond motor controlinto tasks focusing on perceptual processing or sensorimotor learning. Aswith motor function, these nonmotor tasks depend on the cerebellumwhen the task requires the precise representation of temporal information.This more cognitive view of the cerebellum still remains grounded in itsprominent role in the motor system.

Third, within the cerebellum, time is represented in a distributed man­ner, with the exact elements required varying from task to task. The cerebcl-

CEREBELLAR. TIMING S\SfEMS 571

tum performs its temporal computations whenever needed. This may in­clude the programming of the temporal aspects for potential movements,regardless of whether that particular movement is produced.

To say that the cerebellum is involved in cognition does not propel ourunderstanding of the system very far. We need to have specific ideas aboutwhat this contribution might be. This contribution may be timing, buttiming in its many manifestations. This working hypothesis offers a concreteand testable idea about the role of the cerebellum in action, perception,and learning.

The author is grateful to Steve Keele and Scott Grafton for their comments on preliminaryversions of this chapter. The preparation of this chapter wassupported byNIH Grant NS30256.

References

Bartha. G. T., Thompson, R F., and Gluck, M. A. (1992). Sensorimotor learning and thecerebellum. In "Visual Structures and Integrated Functions" (M. Arbib and J. Ewert,eds.). Spring-Verlag, Berlin.

Buonamano, D., and Mauk, M. (1994). Neural network model of the cerebellum: Temporaldiscrimination and the timing of motor responses. Neur. Comput. 6, 38-55.

Daum, I., Schugens, M., Ackermann, H., Lutzenberger, W., Dichgans,]., and Birbaumer,N. (1993). Classical conditioning after cerebellar lesions in humans. Behav. Neurosci.107, 748-756.

Desmondvj., and Moore,]. (1988). Adaptive timing in neural networks: The conditionedresponse. Bioi. Cyber. 58, 405-415.

Deiber, M. P., Passingham, R E., Colebatch.], G., Friston, K. L. Nixon, P. D., and Frackowiak,R. S.]. (1991). Conical areas and the selection of movement: A study with positronemission tomography. Exp. Brain Res. 84, 393-402.

Franz, E., Eliassen.j., Ivry,R, and Gazzeniga, M. (1996a). Dissociation of spatial and temporalcoupling in the bimanual movements of collosotomy patients. Psychol. Sci. 7, 306-310.

Franz, E., Ivry, R, and Helmuth, L. (1996h). Reduced timing variability in patients withunilateral cerebellar lesions during bimanual movements. J. Cogn. Neurosci. 8, 107-118.

Friston, K.]., Frith, C. D., Passingham, R. E., Liddle, P. F., and Frackowiak, R. S.]. (1992).Motor practice and neurophysiological adaptation in the cerebellum: A positron tomogra­phy study. Proc. R. Soc. Lond. B 248, 223-228.

Ghez, C. (1991). The Cerebellum. In "Principles of Neural Science" (E. Kandel,]. Schwartz,and T. Jessell, eds.) , pp. 626-646. Elsevier, New York.

Gibbon,J, and Church, R (1990). Representation of time. CognitionS7, 23-54.Grill, S. E., Hallett, M., Marcus, C., and McShane, L. (1994). Disturbances of kinaesthesia in

patients with cerebellar disorders. Brain 117, 1433-1447.

572 RICHARD lVRY

Grossberg,S., and Schmajuk, N. A (1989). Neural dynamics of adaptive timing and temporaldiscrimination during associative learning. Neur. NetuJarlls 2, 79-102.

Hallett, M., Shahani, B., and Young, R. (1975). EMG analysis of patients with cerebellarlesions.]. Neurol. Neurosurg. Psy,hWt. 58, 1163-1169.

Hayes, A., Davidson, M., Keele, S., and Rafal, R. (1995). Toward a functional analysis of thebasal gangha. Tn Cognitive Neuroscience Society 2nd annual meeting.

Heilman, K. M., Rothi, L. J., and Valenstein, E. (1981). Two forms of ideomotor apraxia.Neurorogy S2, 342-346.

Helmuth, L, and Ivry, R. (1996). When two hands are better than one: Reduced timingvariability during bimanual movements.J Exp.PS'jawL Hum. Percept. Perform. 22, 278-293.

Helmuth, L., and Ivry, R. (1997). The neuropsychology of sequential behavior. In "TheNeuropsychology of Movement" (R. Brown, ed.), in press.

Helmuth, L. L., Ivry, R. B., and Shimizu, N. (1997). Preserved performance by cerebellarpatients on tests ofword generation, discrimination learning, and attention. Leam. Mem.,in pres.~.

Holmes, G. (1939). The cerebellum of man. Brain62,1-30.Hore.j., Wild, B., and Diener, H. (1991). Cerebellar dysmetria at the elbow, wrist, and fingers .

./. NeurofJhy.liol. 65, 563-571.Ivry, R. (1993). Cerebellar involvement in the explicit representation of temporal information.

in "Temporal Information Processing in the Nervous System: Special Reference to Dys­lexia and Dysphasia" (P. Tallal, AM. Galaburda, R. R. Llinas, C. von Euler, eds.) , Vol.682, pp. 214-230. New York Academy of Sciences, New York.

Ivry, R., and Diener, II. C. (1991). Impaired velocity perception in patients with lesions ofthe cerebellum. J. Cogn. Neurosa. S, 355-366.

Ivry, R. B., and Hazeltine, R. E. (1995). The perception and production of temporal intervalsacross a range of durations: Evidence for a common timing mechanism. J Exp. PsydwLHum. 1'l'!Tr~t. 1'1orm. 21, 1-12.

Ivry, R. B., and Keele, S. W. (1989). Timing functions of the cerebellum. J. Cogn. Neurost:i.1, 136-152.

Ivry, R. B., Keele, S. W., and Diener, H. C. (1988). Dissociation of the lateral and medialcerebellum in movement timing and movement execution. Exp. Brain Res. 7S, 167-180.

Jenkins, I. H., Brooks, D.J., Nixon, P. D., Frackowiak. R. 5.]., and Passingham, R. E. (1994).Motor sequence learning: A study with positron emission. J Neumsa. 14(6), 3775-37!10.

Jueptner, M., Rijnqes, M., Weiller, C., Fais.~,J. H., Timmann, D., Mueller, S. P., and Diener,H. C. (1995). Localization of a cerebellar timining process using PET. Neurorogy 45, 1540­1545.

Keele, S. W., and Ivry, R. (1991). Does the cerebellum provide a common computation fordiverse tasks? In "The Development and Neural Bases of Higher Cognitive Functions"(A. Diamond, ed.), Vol. 608, pp. 179-211. New York Academy of Sciences, New York.

Keele, S. W., Pokorny.R, Corcos, D., and Ivry,R. (1985). Do perception and motor productionshare common timing mechanisms? Acta PS'jc.hol. 60, 173-l!13.

Kehoe, E.J., Horne, P. 5., and Horne, A. .J. (1993). Discrimination learning using differentC::SUS intervals in classical conditioning of the rabbit's nictitating membrane response.PS'jehobiolbgy 21, 277-285.

Kim, S., Ugurbil, K., and Strick, P. (1994). Activation of a cerebellar output nucleus duringcognitive processing, Slimee 265, 949-951.

Lavond, D., Lincoln, D., McCormick, D., and Thompson, R. (1984). Effect or bilateral lesionsof the dentate and interpositus nuclei in conditioning of heart rate and nictitatingmembrane/eyelid responses in the rabbit. Brain Res. 305, 323-330.

CEREBELlAR TIMING SYSTEMS 573

Leigh. j., and Zee, D. (1991). "The Neurology of Eye Movements (2nd Ed.). Davis, Phila­delphia.

Nawrot, M., and Rizzo,M. (1995). Motion perception deficits from midline cerebellar lesionsin human. Vision Res. 35, 723-731.

Pentland, A. (1980). Maximum likelihood estimation: The best PEST. Percept. Psychophys.28, 377-379.

Perrett, S., Ruiz,B.,and Mauk, M. (1993). Cerebellarcortex lesions disrupt learning-dependenttiming of conditioned eyelid responses.) Neurosa. 13, 1708-1718.

Petersen, S. E., Fox, P. T., Posner, M. i., Mintun, M., and Raichle, M. E. (1988). Positronemission tomographic studies of the cortical anatomy of single-word processing. Nature331,585-589.

Raichle, M. E., Fiel, j. A., Videen, T. 0., MacLeod, A. K., Pardo,]. V., Fox, P. T., andPetersen, S.E. (1994). Practice-related changes in human brain functional anatomy duringnonmotor learning. Cereb. Cortex. 4, 8-26.

Rizzolatti, G., Gentilucci, M., Camarda, R., Gallese, V., Luppino, G., Matelli, M., and Fogassi,L. (1990). Neurons related to reaching-grasping arm movements in the rostral part ofarea 6. Exp. Brain Res. 82, 337-350.

Robertson, C., and Flowers, K (1990). Motor set in Parkinson's disease.} NfflTOI. NfflTOSUTg.Psychiat. 53, 583-592.

Sergent, V., Hellige, j., and Cherry, B. (1993). Effects of responding hand and concurrentverbal processing on time-keeping and motor-implementation processes. Brain Cogn.23, 243-262.

Thompson, R. (1990). Neural mechanisms of classical conditioning in mammals. Phil. Trans.R. s« (Sect. B) 329, 161-170.

Topka, H., Valls-Sole,j., Massaquoi, S., and Hallett, M. (1993). Deficit in classical conditioningin patients with cerebellar degeneration. Brain 116, 961-969.

Wickens, D. D., Nield, A. F., Tuber, D. 5., and Wickens, C. (1969). Strength, latency, andform of conditioned skeletal and autonomic responses as functions of ('SUS intervals.l Exp. Psychol.80, 165-170.

Wing, A. (1980). The long and short of timing in response sequences. In "Tutorials in MotorBehavior" (G. Stelmach and]. Requin, eds.), North-Holland, New York.

Wing, A., and Kristofferson, A. (1973). Response delays and the timing of discrete motorresponses. Percept. Psychophys. 14,5-12.

Woodruff-Pak, D. 5., Papka, M., and lVI)', R. B. (1996). Cerebellar involvement in eyeblinkclassical conditioning in humans. NfflTopsychoiogy

Yeo, C. H. (1991). Cerebellum and classical conditioning of motor responses. In "Activity­Driven CNS Changes in Learning and Development" O. R. Wolpaw,J. T. Schmidt, andT. M. Vaughan, eds.) , Vol. 627, pp. 292-304. New YorkAcademy of Sciences, New York.